Search papers, labs, and topics across Lattice.
Shanghai Jiao Tong University
3
0
6
Cinematic multi-shot video generators are surprisingly poor editors: aesthetic visual quality correlates only weakly with a model's actual ability to follow fundamental cinematic grammar like J-cuts, transition timing, and montage.
Achieve more reliable ADAS by using a physics-informed neural network that leverages damper characteristics to estimate wheel load more accurately than existing methods.
Achieve better video editing without retraining by dynamically locking background features based on a "hallucination metric" that detects when the diffusion model is about to go astray.
